Prompt Engineering for Paralegals: Tactical Guide

Master prompt engineering for legal work: templates and techniques for paralegals to get better AI outputs on research, drafting, and document review tasks.

The Adaptist Group January 15, 2026 6 min read AI-researched & drafted · Human-edited & fact-checked
person using laptop computer | Photo by Power Digital Marketing on Unsplash
person using laptop computer | Photo by Power Digital Marketing on Unsplash

The difference between a mediocre AI output and a useful one often comes down to how you ask. In 2026, prompt engineering isn’t a technical skill—it’s a communication skill that paralegals can master to dramatically improve AI-assisted legal work.

Why Prompting Matters for Legal Work

AI assistants are powerful but literal. Ask vaguely, get vague results. Ask precisely, get precise results. For legal work—where precision is everything—this difference is critical.

Consider two ways to ask for case research:

Weak Prompt

”Find cases about employment discrimination.”

Result: Generic overview with irrelevant cases

Strong Prompt

”Find Ninth Circuit cases from 2020-2024 where an employer’s failure to provide reasonable accommodation for a mental health condition resulted in a verdict for the plaintiff. Focus on cases involving remote work requests.”

Result: Directly relevant case law

The Anatomy of an Effective Legal Prompt

1. Context Setting

Tell the AI what role it’s playing and what expertise to apply:

“You are a legal research assistant helping prepare a motion for summary judgment. The client is the defendant in a breach of contract case in California state court.”

2. Specific Task

State exactly what output you need:

“Identify the elements of a promissory estoppel claim under California law, and for each element, provide one case citation with a brief quote from the holding.”

3. Constraints

Define boundaries that improve relevance:

“Only cite California Court of Appeal or Supreme Court cases. Only include cases decided after 2015. Do not include federal cases.”

4. Format Requirements

Specify how you want the output structured:

“Format your response as a numbered list. For each case, provide: (1) full citation, (2) one-sentence summary of relevant holding, (3) key quote with page number.”

Prompting Techniques for Common Legal Tasks

Document Summarization

Template

”Summarize this [document type] in [number] bullet points. Focus on: [specific aspects]. Flag any [specific issues to watch for]. The summary will be used for [purpose].”

Example: “Summarize this commercial lease agreement in 10 bullet points. Focus on: rent escalation provisions, termination rights, and tenant improvement allowances. Flag any provisions that deviate from our client’s standard terms. The summary will be used for partner review before signing.”

Case Analysis

Template

”Analyze [case citation] for its application to a situation where [brief fact pattern]. Identify: (1) the key holding, (2) how the court’s reasoning would apply to our facts, (3) any distinguishing factors that could limit its precedential value.”

Contract Clause Drafting

Template

”Draft a [clause type] for inclusion in a [contract type] governed by [jurisdiction] law. The clause should [objective]. It must be consistent with [specific requirements]. Provide three alternative versions: aggressive (favoring our client), neutral, and conservative.”

Discovery Response Drafting

Template

”Draft responses to the following interrogatories. For each response, include appropriate objections based on [grounds]. Where objections are made, also provide a response ‘subject to and without waiving’ the objection. Our client’s position is [brief summary].”

Advanced Techniques

Few-Shot Prompting

Provide examples of the output style you want:

“Write case summaries in the following format:

Example: ‘In Smith v. Jones, 123 Cal.App.4th 456 (2020), the court held that [holding]. This case is relevant because [relevance]. Key quote: [quote].’

Now summarize the following three cases using this format:“

Chain-of-Thought Prompting

Ask the AI to show its reasoning step by step:

“Analyze whether our client has a viable claim for tortious interference. Work through this step by step:

  1. First, identify the elements of tortious interference under California law

  2. Then, analyze each element against these facts: [facts]

  3. Finally, provide your overall assessment with confidence level”

Chain-of-thought prompting produces more reliable results because it forces the AI to engage in systematic analysis rather than jumping to conclusions.

Role-Based Prompting

Have the AI adopt different perspectives:

“First, as plaintiff’s counsel, identify the three strongest arguments for liability. Then, as defense counsel, identify the three strongest counterarguments. Finally, assess which side has the stronger position and why.”

Common Mistakes to Avoid

1. Over-Reliance on AI Output

Never copy AI output directly into a filing. AI is a research and drafting assistant, not a final authority. Always verify citations and review for accuracy. Our guide to AI auditing for paralegals covers systematic methods for catching AI errors before they reach attorneys.

2. Vague Jurisdiction References

Bad: “What’s the law on this?”
Good: “What is the standard under Texas state law for…“

3. Missing Time Constraints

Legal research needs temporal boundaries. “Recent cases” means different things to humans and AI. Specify: “Cases decided between January 2022 and present.” For a broader look at verifying AI-generated research, see our AI fact-checking workflow guide.

4. Asking for Legal Conclusions

AI cannot provide legal advice or predict case outcomes. Ask for research, analysis, and drafting assistance—not conclusions like “will we win?”

Prompt Library: Copy and Customize

Legal Research

Find [number] cases from [jurisdiction] decided after [date] addressing [legal issue]. For each case, provide: full Bluebook citation, court, year, one-paragraph summary of relevant holding. Exclude cases that have been overruled.

Contract Review

Review this [contract type] and identify: (1) provisions that deviate from market standard terms, (2) ambiguous language that could lead to disputes, (3) missing provisions that should be included, (4) provisions that favor the other party. Format as a table with columns: Provision, Issue, Recommendation.

Deposition Prep

Based on the following complaint allegations and discovery responses, generate [number] deposition questions for [witness name/role]. Organize questions by topic. Include follow-up questions for anticipated responses. Focus on [specific issues].

The Verdict

Prompt engineering is the lever that makes AI useful for legal work. Paralegals who master these techniques will produce better work product in less time—and become indispensable as firms increase their AI adoption.

Start with the templates above, customize them for your practice area, and build a personal prompt library. The investment in learning precise prompting pays dividends on every AI interaction.

Frequently Asked Questions

Should I tell the AI I’m a paralegal?

Yes, context helps. Stating your role helps the AI calibrate its response appropriately. “I’m a litigation paralegal preparing materials for attorney review” sets useful expectations about the level of detail and verification needed.

How long should prompts be?

As long as necessary, but no longer. A complex research task might need 200+ words of context. A simple formatting request might need 20 words. Err on the side of more detail—AI won’t be overwhelmed, and vagueness is the main source of poor outputs.

Can I use the same prompts with different AI tools?

Mostly yes. Well-structured prompts work across Claude, ChatGPT, Copilot, and legal-specific tools. Some tools have specific features (like document upload) that may require adaptation, but the core principles are universal.

More in Career & Skills

From Other Topics